This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: mknumeric_limits



I trust Benjamin and Gabriel will correct any mistakes I'm about to make. :-)

On Mon, Feb 19, 2001 at 10:01:44AM -0700, Jeffrey A Law wrote:
> First, it seems rather odd to be using "uname" like that, unless this code
> is supposed to be run on the build machine.
> 
> Second, by using uname instead of a triplet, we lose the ability to
> select behavior on the triplet.  This is important when you consider
> systems like the PA which have two completely different toolchains that
> can not interoperate with each other.

That fragment predates just about everything else, such as configure.host
and configure.target, which are recent and use triplets.  mknumeric_limits
could be changed to use those, or maybe could call the generated tests_flags
script.  (Gaby? Sanity check?)


> Third, what's the purpose behind -nodefaultlibs, then listing the libraries
> you want explicitly?  If we're really just building something for the
> build system, why not let the compiler figure out what libraries need to
> be linked to create an executable?

Because later in that script we call the C++ compiler, which by default
will link in the c++ library... which we haven't built yet.  (There's code
like this in a couple of places, like src/Makefile.am.  Arguably we could
call libtool in mknumeric_limits and have it link gen_num_limits.)


Phil

-- 
pedwards at disaster dot jaj dot com  |  pme at sources dot redhat dot com
devphil at several other less interesting addresses in various dot domains
The gods do not protect fools.  Fools are protected by more capable fools.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]